CN110101362B - Method for removing image noise related to OCT and OCTA - Google Patents

Method for removing image noise related to OCT and OCTA Download PDF

Info

Publication number
CN110101362B
CN110101362B CN201910352015.7A CN201910352015A CN110101362B CN 110101362 B CN110101362 B CN 110101362B CN 201910352015 A CN201910352015 A CN 201910352015A CN 110101362 B CN110101362 B CN 110101362B
Authority
CN
China
Prior art keywords
image
frequency domain
octa
denoised
sub
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201910352015.7A
Other languages
Chinese (zh)
Other versions
CN110101362A (en
Inventor
杨建龙
方黎洋
王浩
郭雨荟
胡衍
刘江
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ningbo Institute of Material Technology and Engineering of CAS
Cixi Institute of Biomedical Engineering CIBE of CAS
Original Assignee
Ningbo Institute of Material Technology and Engineering of CAS
Cixi Institute of Biomedical Engineering CIBE of CAS
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ningbo Institute of Material Technology and Engineering of CAS, Cixi Institute of Biomedical Engineering CIBE of CAS filed Critical Ningbo Institute of Material Technology and Engineering of CAS
Priority to CN201910352015.7A priority Critical patent/CN110101362B/en
Publication of CN110101362A publication Critical patent/CN110101362A/en
Application granted granted Critical
Publication of CN110101362B publication Critical patent/CN110101362B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B3/00—Apparatus for testing the eyes; Instruments for examining the eyes
    • A61B3/10—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions
    • A61B3/102—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions for optical coherence tomography [OCT]
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B3/00—Apparatus for testing the eyes; Instruments for examining the eyes
    • A61B3/10—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions
    • A61B3/12—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions for looking at the eye fundus, e.g. ophthalmoscopes
    • A61B3/1225—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions for looking at the eye fundus, e.g. ophthalmoscopes using coherent radiation
    • A61B3/1233—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions for looking at the eye fundus, e.g. ophthalmoscopes using coherent radiation for measuring blood flow, e.g. at the retina
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B3/00—Apparatus for testing the eyes; Instruments for examining the eyes
    • A61B3/10—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions
    • A61B3/12—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions for looking at the eye fundus, e.g. ophthalmoscopes
    • A61B3/1241—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions for looking at the eye fundus, e.g. ophthalmoscopes specially adapted for observation of ocular blood flow, e.g. by fluorescein angiography
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B3/00—Apparatus for testing the eyes; Instruments for examining the eyes
    • A61B3/10—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions
    • A61B3/14—Arrangements specially adapted for eye photography
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation
    • G06T11/003—Reconstruction from projections, e.g. tomography
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/70—Denoising; Smoothing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10101—Optical tomography; Optical coherence tomography [OCT]

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Public Health (AREA)
  • Veterinary Medicine (AREA)
  • Biophysics (AREA)
  • Ophthalmology & Optometry (AREA)
  • General Health & Medical Sciences (AREA)
  • Animal Behavior & Ethology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Medical Informatics (AREA)
  • Molecular Biology (AREA)
  • Surgery (AREA)
  • Hematology (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Vascular Medicine (AREA)
  • Image Processing (AREA)
  • Image Analysis (AREA)

Abstract

The invention belongs to the technical field of medical image processing, and particularly relates to an image noise removing method for OCT (optical coherence tomography) and OCTA (optical coherence tomography), which is used for eyeball micro-motion noise in en face images of the OCT and OCTA. The method comprises the following steps: step S1, extracting a frequency domain image; step S2, removing stripe information in the frequency domain image to obtain a de-noised frequency domain image; and step S3, reconstructing a denoised image based on the denoised frequency domain image. The eyeball micromotion noise is a bright short line in the en face image in the horizontal direction, and the horizontal micromotion artifact can be extracted from the image in the two-dimensional transform domain. According to the technical scheme, the face OCTA image is converted or decomposed to obtain a frequency domain image, then the frequency domain image is denoised by a filter, and finally the noise-removed angiography image is reconstructed based on the denoised frequency domain image.

Description

Method for removing image noise related to OCT and OCTA
Method for removing image noise related to OCT and OCTA
Technical Field
The invention belongs to the technical field of medical image processing, and particularly relates to an image noise removing method for OCT (optical coherence tomography) and OCTA (optical coherence tomography), which is used for eyeball micro-motion noise in en face images of the OCT and OCTA.
Background
Optical Coherence Tomography Angiography (OCTA) is a non-invasive angiographic imaging technique that has been widely used in recent years for the study and diagnosis of diseases in the fundus region. However, OCTA requires a longer acquisition duration than photography-based modes (such as slit lamps and fundus cameras). Domestic and foreign research shows that human eyes are in a micro-motion state when observing scenes, and the micro-motion has three modes: high frequency flutter, drift motion and flicker. Therefore, the micro-motion of the eyeball during the acquisition process becomes a main source of image artifacts, and the generated horizontal bright short lines can seriously affect the accuracy of quantitative analysis and medical diagnosis.
At present, great efforts have been made to eliminate motion artifacts in the OCTA on the hardware systems of the OCTA. For example, an invention patent granted publication No. CN102085093B, publication No. 2013, 11/27, relates to an image processing apparatus for processing a tomographic image of an eye to be examined, which generates a high-resolution low-noise tomographic image while minimizing the influence of an eye-jiggling, a head movement, and the like. A detection unit for detecting a movement amount of an eye by using a signal obtained by photographing a tomographic image, and a determination unit for determining a scanning beam used for photographing the tomographic image based on the movement amount detected by the detection unit are included.
The method of combining the acquisition system of the OCTA with the eye tracking mode (such as Scanning Laser Ophthalmoscope (SLO)) can effectively reduce the blood vessel discontinuity and the volume motion noise, which has become a standard set of most commercial OCTA systems. However, ocular twitter noise in the form of bright dashes along the fast scan direction still appears on many clinical datasets. Therefore, the extraction and quantification of blood vessel related features may be affected by such noise, which further introduces inaccuracies to the analysis and diagnosis of eye diseases.
Unlike isotropic and randomly distributed noise, such as pepper noise and speckle noise, eye jiggle noise has strong directivity, and thus may not be suitable for conventional denoising methods, such as Kalman filtering. Fourier Transform (FT) is the transformation of an image from the spatial domain into the frequency domain, and then the required processing of the image in the frequency domain with the relevant filters. Fourier transforms can address many image processing requirements with their time and frequency domain methods. Wavelet Transform (WT) is variable in frequency and position in terms of image denoising, sometimes with frequency locality. After wavelet transform, the image is mainly concentrated on a few wavelet coefficients with larger absolute amplitudes, and noise is scattered on some wavelet coefficients with smaller absolute amplitudes, so that the wavelet coefficients can be denoised by utilizing a shrinkage threshold, and the purpose of denoising is achieved.
The Contourlet transform is a multi-scale geometric analysis tool, is a two-dimensional representation method of images in the true sense, has good multi-resolution, localization, directivity and other excellent characteristics, extends the advantages of wavelets to a high-dimensional space, can better depict the characteristics of high-dimensional information, and is more suitable for processing the information with hyperplane singularity. The non-subsampled contourlet transform (NSCT) is a non-subsampled, multi-scale transform with translational invariance, and its anisotropic contourlet basis gives NSCT the line singularity advantage of characterizing images. The NSCT can provide more abundant time domain information and accurate frequency localization information, elements in each image sub-band coefficient are in one-to-one correspondence with pixels in an image space domain, edge information in the image space domain is easily and directly detected by using the distribution rule of the NSCT domain coefficient, and operations such as reconstruction are not needed.
Disclosure of Invention
The invention provides an OCT (optical coherence tomography) and OCTA (optical coherence tomography) image noise removal method based on WT multi-scale advantages, NSCT multi-directional advantages and a Fourier filter.
A method of image noise removal for OCT and OCTA comprising:
step S1, extracting a frequency domain image;
step S2, removing stripe information in the frequency domain image to obtain a de-noised frequency domain image;
and step S3, reconstructing a denoised image based on the denoised frequency domain image.
The eyeball micromotion noise is a bright short line in the en face image in the horizontal direction, and the horizontal micromotion artifact can be extracted from the image in the two-dimensional transform domain. According to the technical scheme, the face OCTA image is converted or decomposed to obtain a frequency domain image, then the frequency domain image is denoised by a filter, and finally the noise-removed angiography image is reconstructed based on the denoised frequency domain image.
Preferably, the step S1 includes a step S1-1 of performing a two-dimensional fourier transform. To convert the image from the spatial domain to the frequency domain.
Preferably, the step S1-1 is preceded by a step S1-0 of extracting a horizontal sub-band image containing stripe information; in step S1-1, the horizontal subband image is subjected to two-dimensional fourier transform. And extracting a horizontal sub-band image containing stripe information, performing two-dimensional Fourier transform, and then performing frequency domain denoising, so that useful information in an original image can be better reserved.
Preferably, the wavelet decomposition with the maximum number of times of L is performed in the step S1-0
Figure 100002_DEST_PATH_IMAGE002
Wherein
Figure 100002_DEST_PATH_IMAGE004
(ii) a In the step S1-1, the horizontal sub-band image containing the stripe information is processed
Figure 100002_DEST_PATH_IMAGE006
And carrying out two-dimensional Fourier transform to obtain the frequency domain image. Wavelet transformation and Fourier filter are combined to construct a fast, strong and stable filter for eliminating horizontal stripes of images, and the micromotion artifacts and the original characteristics are strictly separated, so that not only can unnecessary structures be inhibited, but also the information of the original images can be highly stored;
preferably, the step S1-0 includes: step S1-0-1, performing nonsubsampled contourlet decomposition to obtain a high-pass sub-band image and a low-pass sub-band image; step S1-0-2, the low pass sub-band image is processedDecomposing the non-downsampling contourlet for m times to obtain m high-frequency sub-band images; step S1-0-3, performing non-down sampling directional filter bank decomposition on the m high frequency sub-band images to obtain horizontal sub-band images
Figure 100002_DEST_PATH_IMAGE008
And vertical subband images
Figure 100002_DEST_PATH_IMAGE010
Wherein
Figure 100002_DEST_PATH_IMAGE012
(ii) a In the step S1-1, the horizontal sub-band image containing the stripe information is processed
Figure 100002_DEST_PATH_IMAGE014
And carrying out two-dimensional Fourier transform to obtain the frequency domain image. The non-downsampled contourlet domain has the best performance in terms of noise cancellation and vessel preservation.
Preferably, the step S2 includes: step S2-1, the zero frequency point of the frequency domain image is moved to the middle of the frequency spectrum; step S2-2, multiplying by a Gaussian damping function; and S2-3, restoring the zero frequency point.
Preferably, in step S3, the denoised image is obtained by performing two-dimensional inverse fourier transform.
Preferably, the step S3 includes: step S3-1, performing two-dimensional inverse Fourier transform to obtain a denoised sub-band image
Figure 100002_DEST_PATH_IMAGE016
(ii) a Step S3-2, wavelet reconstruction is carried out
Figure 100002_DEST_PATH_IMAGE018
And obtaining the denoised image.
Preferably, the step S3 includes: step S3-1, performing two-dimensional inverse Fourier transform to obtain a denoised sub-band image
Figure 100002_DEST_PATH_IMAGE020
(ii) a Step S3-2, based on the denoised sub-band image
Figure DEST_PATH_IMAGE022
And performing non-subsampled contourlet reconstruction to obtain the denoised image. The frequency domain denoising processing of the Fourier filter can keep the sharpness of the image edge; the two-dimensional transform domain Fourier filter can effectively eliminate eyeball micro-motion noise of the en face image; the two-dimensional transform domain Fourier filter provides a better visualization effect for the en face image object, and does not introduce additional artifacts into the restored image; two-dimensional transform domain fourier filter techniques would be beneficial for quantification of vessel-related analysis and diagnosis.
Preferably, the step S1 is preceded by an image preprocessing step S0, and the step S0 includes: step S0-1, converting the image into a gray image; and step S0-2, normalization processing.
The invention has the following beneficial effects:
1. the frequency domain denoising processing of the Fourier filter can keep the sharpness of the image edge;
2. the two-dimensional transform domain Fourier filter can effectively eliminate eyeball micro-motion noise of the en face image;
3. the two-dimensional transform domain Fourier filter provides a better visualization effect for the en face image object, and does not introduce additional artifacts into the restored image;
4. wavelet transformation and Fourier filter are combined to construct a fast, strong and stable filter for eliminating horizontal stripes of images, and the micromotion artifacts and the original characteristics are strictly separated, so that not only can unnecessary structures be inhibited, but also the information of the original images can be highly stored;
5. the non-downsampling contourlet domain has the best performance in the aspects of noise elimination and blood vessel preservation;
6. two-dimensional transform domain fourier filter techniques would be beneficial for quantification of vessel-related analysis and diagnosis.
Drawings
FIG. 1 illustrates the steps of the imaging method of the present invention;
fig. 2 is a comparison of a cross-sectional image of raw data and a cross-sectional image of preprocessed data according to a first embodiment of the present invention.
Detailed Description
The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. Unless otherwise defined, all terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It will be further understood that the conventional terms should be interpreted as having a meaning that is consistent with their meaning in the relevant art and this disclosure. The present disclosure is to be considered as an example of the invention and is not intended to limit the invention to the particular embodiments.
Example one
A method for denoising an en face image of OCTA (optical clear associated with a transform-domain Fourier transform) by using a two-dimensional transform-domain Fourier filter is based on Fourier Transform (FT) denoising, and specifically comprises the following steps:
step S0, image preprocessing step. And preprocessing the en face image of the OCTA so as to facilitate subsequent unified processing by adopting a standard method. The method comprises the following steps:
step S0-1, converting the en face image of the OCTA into a gray level image;
step S0-2, normalizing the gray level image to obtain a gray level image
Figure DEST_PATH_IMAGE024
。
In step S1, a frequency domain image is extracted from the image obtained in step S0. In this embodiment, for gray scale images
Figure DEST_PATH_IMAGE026
Performing two-dimensional Fourier transform to realize gray image
Figure DEST_PATH_IMAGE028
And converting from a space domain to a frequency domain to obtain a frequency domain image.
And step S2, removing the stripe information in the frequency domain image obtained in the step S1, and further obtaining a de-noised frequency domain image. The method can be realized by adopting a frequency domain filtering method in the prior art, such as Fourier filtering, wiener filtering, adaptive filtering and the like. The method is preferably realized by adopting a Fourier filter, the frequency domain denoising treatment of the Fourier filter can keep the sharpness of the edge of the image, the two-dimensional transform domain Fourier filter can effectively eliminate the micro-eye noise of the en face image, a better visualization effect is provided for the en face image object, no extra artifact is introduced into the recovered image, and the method is beneficial to the quantification of the blood vessel correlation analysis and diagnosis. The method specifically comprises the following steps:
and step S2-1, shifting the zero frequency point of the frequency domain image to the middle of the frequency spectrum so as to further tighten the stripe information into a narrow band in the vertical direction in the frequency domain image.
Step S2-2, the frequency domain image obtained in the step S2-1 and a Gaussian damping function
Figure DEST_PATH_IMAGE030
The multiplication eliminates the stripe information compressed in this way.
And S2-3, restoring the zero frequency point of the frequency domain image obtained in the step S2-2 to obtain a de-noised frequency domain image.
And S3, reconstructing a denoised image based on the denoised frequency domain image obtained in the S2. In this step, a corresponding method is selected for image reconstruction according to the method for extracting the frequency domain image used in step S1. In this embodiment, in step S1, a frequency domain image is obtained by fourier transform, and correspondingly, in step S3, a denoised image is obtained by performing inverse fourier transform
Figure DEST_PATH_IMAGE032
。
Example two
A method for denoising an en face image of OCTA (optical clear associated with transform-domain transform) by using a two-dimensional transform domain Fourier filter is based on Wavelet Transform (WT) for denoising, and specifically comprises the following steps:
step S0, image preprocessing step. And preprocessing the en face image of the OCTA so as to facilitate subsequent unified processing by adopting a standard method. The method comprises the following steps:
step S0-1, converting the en face image of the OCTA into a gray level image;
step S0-2, normalizing the gray level image to obtain a gray level image
Figure DEST_PATH_IMAGE034
。
In step S1, a frequency domain image is extracted from the image obtained in step S0. In the present embodiment, first, in step S1-0, the gradation image obtained in step S1 is subjected to
Figure DEST_PATH_IMAGE036
Decomposing to obtain a horizontal sub-band image containing stripe information; and then, the horizontal subband image obtained in the step S1-0 is subjected to two-dimensional Fourier transform in a step S1-1 to obtain a frequency domain image. The embodiment adopts wavelet transform to perform decomposition, and specifically comprises the following steps:
step S1-0, for gray scale image
Figure DEST_PATH_IMAGE038
Wavelet decomposition
Figure DEST_PATH_IMAGE040
(wherein,
Figure DEST_PATH_IMAGE042
l is the maximum number of wavelet decompositions), thereby obtaining a gray-scale image
Figure DEST_PATH_IMAGE044
The structural information contained in (A) is divided into horizontal sub-band images at different resolution scales
Figure DEST_PATH_IMAGE046
Vertical subband images
Figure DEST_PATH_IMAGE048
And diagonal detail sub-band images
Figure DEST_PATH_IMAGE050
。
Step S1-1, for horizontal subband image containing stripe information
Figure DEST_PATH_IMAGE052
And performing two-dimensional Fourier transform to obtain a frequency domain image.
And step S2, removing the stripe information in the frequency domain image obtained in the step S1, and further obtaining a de-noised frequency domain image. The method can be realized by adopting a frequency domain filtering method in the prior art, such as Fourier filtering, wiener filtering, adaptive filtering and the like. The method is preferably realized by adopting a Fourier filter, the frequency domain denoising treatment of the Fourier filter can keep the sharpness of the edge of the image, the two-dimensional transform domain Fourier filter can effectively eliminate the micro-eye noise of the en face image, a better visualization effect is provided for the en face image object, no extra artifact is introduced into the recovered image, and the method is beneficial to the quantification of the blood vessel correlation analysis and diagnosis. The method specifically comprises the following steps:
and step S2-1, shifting the zero frequency point of the frequency domain image to the middle of the frequency spectrum so as to further tighten the stripe information into a narrow band in the vertical direction in the frequency domain image.
Step S2-2, the frequency domain image obtained in the step S2-1 and a Gaussian damping function
Figure DEST_PATH_IMAGE054
The multiplication eliminates the stripe information compressed in this way.
And S2-3, restoring the zero frequency point of the frequency domain image obtained in the step S2-2 to obtain a de-noised frequency domain image.
And S3, reconstructing a denoised image based on the denoised frequency domain image obtained in the S2. In this step, a corresponding method is selected for image reconstruction according to the method for extracting the frequency domain image used in step S1. In this embodiment, in step S1, a frequency domain image is obtained by wavelet decomposition, and correspondingly, in step S3, a denoised image is obtained by performing inverse wavelet transform
Figure DEST_PATH_IMAGE056
. The method specifically comprises the following steps:
s3-1, performing two-dimensional inverse Fourier transform on the de-noised frequency domain image obtained in the S2 to obtain a de-noised sub-band image
Figure DEST_PATH_IMAGE058
；
Step S3-2, will
Figure DEST_PATH_IMAGE060
Haar wavelet reconstruction is carried out to obtain a de-noised image
Figure DEST_PATH_IMAGE062
。
Wavelet Transform (WT) and Fourier filter are combined to construct a fast, powerful and stable filter for eliminating horizontal stripes of image, and the micromotion artifact and original features are strictly separated, so that not only can unnecessary structures be restrained, but also original image information can be highly stored.
EXAMPLE III
A method for denoising an en face image of OCTA (optical coherence tomography) by using a two-dimensional transform domain Fourier filter is based on non-subsampled contourlet transform (NSCT) for denoising, and specifically comprises the following steps:
step S0, image preprocessing step. And preprocessing the en face image of the OCTA so as to facilitate subsequent unified processing by adopting a standard method. The method comprises the following steps:
step S0-1, converting the en face image of the OCTA into a gray level image;
step S0-2, normalizing the gray level image to obtain a gray level image
Figure DEST_PATH_IMAGE064
。
In step S1, a frequency domain image is extracted from the image obtained in step S0. In the present embodiment, first, in step S1-0, the gradation image obtained in step S1 is subjected to
Figure DEST_PATH_IMAGE066
Decomposing to obtain a horizontal sub-band image containing stripe information; and then, the horizontal subband image obtained in the step S1-0 is subjected to two-dimensional Fourier transform in a step S1-1 to obtain a frequency domain image. This embodiment employs non-downsampling contourlet transformThe decomposition is carried out, and the optimal performance in noise elimination and blood vessel preservation is achieved; the method specifically comprises the following steps:
step S1-0-1, for gray scale image
Figure DEST_PATH_IMAGE068
Performing non-subsampled pyramid (NSP) decomposition to generate a high-pass sub-band image and a low-pass sub-band image, and satisfying the image reconstruction condition:
Figure DEST_PATH_IMAGE070
. Wherein,
Figure DEST_PATH_IMAGE072
in the form of a low-pass filter,
Figure DEST_PATH_IMAGE074
for the high pass filter, set to:
Figure DEST_PATH_IMAGE076
，
Figure DEST_PATH_IMAGE078
and
Figure DEST_PATH_IMAGE080
in order to synthesize the filter, the filter is,
Figure DEST_PATH_IMAGE082
。
and step S1-0-2, continuing to perform non-downsampling pyramid decomposition on the low-pass sub-band images for m times to obtain m high-frequency sub-band images.
Step S1-0-3, performing non-down sampling directional filter bank decomposition on the m high frequency sub-band images to obtain horizontal sub-band images
Figure DEST_PATH_IMAGE084
And vertical subband images
Figure DEST_PATH_IMAGE086
Wherein
Figure DEST_PATH_IMAGE088
。
Step S1-1, for horizontal subband image containing stripe information
Figure DEST_PATH_IMAGE090
And performing two-dimensional Fourier transform to obtain a frequency domain image.
And step S2, removing the stripe information in the frequency domain image obtained in the step S1, and further obtaining a de-noised frequency domain image. The method can be realized by adopting a frequency domain filtering method in the prior art, such as Fourier filtering, wiener filtering, adaptive filtering and the like. The method is preferably realized by adopting a Fourier filter, the frequency domain denoising treatment of the Fourier filter can keep the sharpness of the edge of the image, the two-dimensional transform domain Fourier filter can effectively eliminate the micro-eye noise of the en face image, a better visualization effect is provided for the en face image object, no extra artifact is introduced into the recovered image, and the method is beneficial to the quantification of the blood vessel correlation analysis and diagnosis. The method specifically comprises the following steps:
and step S2-1, shifting the zero frequency point of the frequency domain image to the middle of the frequency spectrum so as to further tighten the stripe information into a narrow band in the vertical direction in the frequency domain image.
Step S2-2, the frequency domain image obtained in the step S2-1 and a Gaussian damping function
Figure DEST_PATH_IMAGE092
The multiplication eliminates the stripe information compressed in this way.
And S2-3, restoring the zero frequency point of the frequency domain image obtained in the step S2-2 to obtain a de-noised frequency domain image.
And S3, reconstructing a denoised image based on the denoised frequency domain image obtained in the S2. In this step, a corresponding method is selected for image reconstruction according to the method for extracting the frequency domain image used in step S1. In this embodiment, in step S1, a frequency domain image is obtained by wavelet decomposition, and correspondingly, in step S3, a denoised image is obtained by performing inverse wavelet transform
Figure DEST_PATH_IMAGE094
. Specifically comprises：
S3-1, performing two-dimensional inverse Fourier transform on the de-noised frequency domain image obtained in the S2 to obtain a de-noised sub-band image
Figure DEST_PATH_IMAGE096
；
Step S3-2, based on the denoised sub-band image obtained in step S3-1
Figure DEST_PATH_IMAGE098
Performing non-down sampling contourlet reconstruction to obtain a denoised image
Figure DEST_PATH_IMAGE100
。
As shown in FIG. 2, 8 typical micro-motion noise regions are selected to show the de-noising results of the above three embodiments of the present invention, and the performance thereof is evaluated and compared. The four regions of interest on the left are used to demonstrate noise removal capability and the other four regions on the right are used to show vessel retention capability. From region one to region four, the NSCT filter used in example three was found to have the best noise removal effect when viewed simply with the naked eye, but was not able to be an accurate conclusion. From region five to region eight, vessel retention is difficult to discern with the naked eye. The comparison was performed by the quantitative indicators shown below. These methods of assessment can be divided into two aspects.
First in terms of the elimination of the jogging noise:
the denoising task typically adds analog noise to the noise-free image, and then compares the denoised image with ground truth for evaluation. However, the source of en face micropower noise is rather complex (decorrelation of time-separated backscattered photons) and there is no suitable modeling method. Thus, the present invention proposes two indicators that do not require a ground truth. First, the entropy of the image, which can be used to represent the amount of information contained in the image, is smaller, the better the noise cancellation effect. Second, the vascular density (vd) should be broadly applied to the quantitative analysis of OCTA images. The eyeball micro-motion noise on the en face image is a bright short line, and the eyeball micro-motion noise is counted as a blood vessel during calculation, so that the blood vessel density is increased.
Second in terms of vessel preservation intact:
the most important thing in this respect is that the noise filter used does not destroy the original shape of the vessel. Therefore, the region of the image with less eye movement noise is used as the reference region. The denoised images are then compared to them. Peak signal-to-noise ratio (PSNR) and Structural Similarity Index (SSIM) were used for evaluation.
Figure DEST_PATH_IMAGE102
TABLE 1
As shown in table 1, table 1 shows a comparison of the above evaluation index to different examples (i.e., different filters). All eight selected regions are used to calculate the index, with the best results in each class shown in bold. In the aspect of removing the eyeball twitter noise, the third embodiment has better performance in both the image entropy and the blood Vessel Density (VD). Example three also has optimal vessel retention on the Structural Similarity Index (SSIM). The fourier filtering of the first embodiment has the highest peak signal-to-noise ratio (PSNR) in vessel retention, but is insufficient in other indexes.
Summary of the experiments:
the method provided by the invention well removes the bright short-line noise generated by eyeball micromotion in an en face image of OCTA by combining Fourier filtering on the basis of fully utilizing the multi-directional advantages of WT and NSCT. In the method, Fourier filtering after non-subsampled contour transform (NSCT) has the best effect on removing the micro-motion noise of the en face image. For vessel retention, wavelet domain filtering has the advantage of preserving signal-to-noise ratio, while NSCT filtering largely preserves structural similarity. The advantages of the invention are illustrated by the objective evaluation of the experimental results.
Although the embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art may make various changes or modifications within the scope of the appended claims.

Claims (8)

1. A method for image noise removal for OCT and OCTA, comprising:
step S0, preprocessing the image to obtain a gray scale image;
step S1, extracting a frequency domain image from the grayscale image;
step S2, removing stripe information in the frequency domain image to obtain a de-noised frequency domain image;
step S3, reconstructing a denoised image based on the denoised frequency domain image;
the step S1 includes:
step S1-0, decomposing the gray level image by adopting non-downsampling contourlet transform or wavelet change to extract a horizontal sub-band image containing stripe information;
and step S1-1, performing two-dimensional Fourier transform on the horizontal sub-band image to obtain a frequency domain image.
2. The method of OCT and OCTA related image noise removal of claim 1, wherein:
in the step S1-0, wavelet decomposition with the maximum number of times of L is performed
Figure DEST_PATH_IMAGE002
Wherein
Figure DEST_PATH_IMAGE004
；
In the step S1-1, the horizontal sub-band image containing the stripe information is processed
Figure DEST_PATH_IMAGE006
And carrying out two-dimensional Fourier transform to obtain the frequency domain image.
3. The method for OCT and OCTA image noise removal according to claim 1, wherein said step S1-0 comprises:
step S1-0-1, performing nonsubsampled contourlet decomposition to obtain a high-pass sub-band image and a low-pass sub-band image; step S1-0-2, performing non-downsampling contourlet decomposition on the low-pass sub-band image for m times to obtain m high-frequency sub-band images;
step S1-0-3, performing non-down sampling directional filter bank decomposition on the m high frequency sub-band images to obtain horizontal sub-band images
Figure DEST_PATH_IMAGE008
And vertical subband images
Figure DEST_PATH_IMAGE010
Wherein;
in the step S1-1, the horizontal sub-band image containing the stripe information is processed
Figure DEST_PATH_IMAGE012
And carrying out two-dimensional Fourier transform to obtain the frequency domain image.
4. The method for image noise removal for OCT and OCTA according to any one of claims 1-3, wherein said step S2 comprises:
step S2-1, the zero frequency point of the frequency domain image is moved to the middle of the frequency spectrum;
step S2-2, multiplying by a Gaussian damping function;
and S2-3, restoring the zero frequency point.
5. The method of OCT and OCTA image noise removal as claimed in claim 4, wherein:
in step S3, two-dimensional inverse fourier transform is performed to obtain the denoised image.
6. The method for OCT and OCTA image noise removal according to claim 5, wherein said step S3 comprises:
step S3-1, performing two-dimensional inverse Fourier transform to obtain a denoised sub-band image
Figure DEST_PATH_IMAGE014
；
Step S3-2, will
Figure DEST_PATH_IMAGE016
And performing wavelet reconstruction to obtain the denoised image.
7. The method for OCT and OCTA image noise removal according to claim 5, wherein said step S3 comprises:
step S3-1, performing two-dimensional inverse Fourier transform to obtain a denoised sub-band image
Figure DEST_PATH_IMAGE018
；
Step S3-2, based on the denoised sub-band image
Figure DEST_PATH_IMAGE020
And performing non-subsampled contourlet reconstruction to obtain the denoised image.
8. The method for OCT and OCTA image noise removal according to claim 1, wherein said step S0 comprises:
step S0-1, converting the image into a gray image;
and step S0-2, normalization processing.
CN201910352015.7A 2019-04-29 2019-04-29 Method for removing image noise related to OCT and OCTA Active CN110101362B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910352015.7A CN110101362B (en) 2019-04-29 2019-04-29 Method for removing image noise related to OCT and OCTA

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910352015.7A CN110101362B (en) 2019-04-29 2019-04-29 Method for removing image noise related to OCT and OCTA

Publications (2)

Publication Number Publication Date
CN110101362A CN110101362A (en) 2019-08-09
CN110101362B true CN110101362B (en) 2021-12-17

Family

ID=67487325

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910352015.7A Active CN110101362B (en) 2019-04-29 2019-04-29 Method for removing image noise related to OCT and OCTA

Country Status (1)

Country Link
CN (1) CN110101362B (en)

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111243087B (en) * 2020-01-20 2023-11-21 上海市第一人民医院 Three-dimensional reconstruction method and device for fundus blood vessel and electronic equipment
CN111861917A (en) * 2020-07-10 2020-10-30 温州医科大学 Choroidal OCT image enhancement method and device based on signal reverse compensation
CN111915515A (en) * 2020-07-21 2020-11-10 无锡声亚医疗科技有限公司 Method for removing noise in ultrasonic image, ultrasonic equipment and storage medium
CN112330547B (en) * 2020-10-12 2024-04-30 佛山科学技术学院 Filtering method for eliminating vascular shadows aiming at OCT retina image
CN113592741A (en) * 2021-08-04 2021-11-02 西北工业大学 Digital image processing method
CN113643396B (en) * 2021-08-05 2023-05-12 中国科学院苏州生物医学工程技术研究所 Swept OCT image processing method and device
CN114037769B (en) * 2021-09-22 2022-12-09 图湃(北京)医疗科技有限公司 Optical coherence tomography angiography method and device, electronic device and storage medium

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103247035A (en) * 2013-05-20 2013-08-14 重庆邮电大学 Medical image processing device, method and system based on digital X-ray machine
CN104282007A (en) * 2014-10-22 2015-01-14 长春理工大学 Contourlet transformation-adaptive medical image fusion method based on non-sampling
CN105184752A (en) * 2015-09-23 2015-12-23 成都融创智谷科技有限公司 Image processing method based on wavelet transform
CN107388963A (en) * 2017-07-13 2017-11-24 北京理工大学 The digital Moiré patterns phase extraction method that wavelet analysis and LPF are combined

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
AU2014202322A1 (en) * 2014-04-29 2015-11-12 Canon Kabushiki Kaisha Wavelet denoising of fringe image
CN103940371A (en) * 2014-05-12 2014-07-23 电子科技大学 High-precision three-dimensional shape measurement method for jump object
CN104580937B (en) * 2015-01-21 2017-06-27 中国科学院上海技术物理研究所 A kind of infrared imaging system fringes noise minimizing technology
CN107228632B (en) * 2017-05-18 2019-12-10 广东工业大学 displacement field chromatographic measurement device and method based on windowed Fourier transform

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103247035A (en) * 2013-05-20 2013-08-14 重庆邮电大学 Medical image processing device, method and system based on digital X-ray machine
CN104282007A (en) * 2014-10-22 2015-01-14 长春理工大学 Contourlet transformation-adaptive medical image fusion method based on non-sampling
CN105184752A (en) * 2015-09-23 2015-12-23 成都融创智谷科技有限公司 Image processing method based on wavelet transform
CN107388963A (en) * 2017-07-13 2017-11-24 北京理工大学 The digital Moiré patterns phase extraction method that wavelet analysis and LPF are combined

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
条纹模式的自适应稀疏滤波方法研究;张峰;《中国优秀硕士学位论文库》;20141015;说明书第2.2.2节 *

Also Published As

Publication number Publication date
CN110101362A (en) 2019-08-09

Similar Documents

Publication Publication Date Title
CN110101362B (en) Method for removing image noise related to OCT and OCTA
KR100646715B1 (en) Method for improving a quality of a 2-d ultrasound image by post-processing
WO2009128213A1 (en) Medical diagnostic device and method of improving image quality of medical diagnostic device
US20070071292A1 (en) Speckle adaptive medical image processing
KR20130007905A (en) Method and apparatus for processing a ultrasound image
US8139891B2 (en) System and method for structure enhancement and noise reduction in medical images
Mitiche et al. Medical image denoising using dual tree complex thresholding wavelet transform
Bin-Habtoor et al. Removal speckle noise from medical image using image processing techniques
Raj et al. Ultrasound medical image denoising using hybrid bilateral filtering
Resham et al. Noise reduction, enhancement and classification for sonar images
Shanmugavadivu et al. Wavelet transformation-based detection of masses in digital mammograms
CN113034374A (en) OCT image adaptive noise estimation method and system
Lazrag et al. Despeckling of intravascular ultrasound images using curvelet transform
Loizou et al. Despeckle filtering of ultrasound images
CN115249235A (en) Method for improving blood vessel information extraction accuracy applied to OCTA in-vivo imaging
Bhairnallykar et al. Pre-processing of multimodal MR images using NLM and histogram equalization
CN111462065B (en) Breast cancer detection method and system by means of ultrasonic and infrared sequence image fusion
CN114387173A (en) OCT image noise reduction method, electronic device and storage medium
Lee et al. Enhancement of blood vessels in retinal imaging using the nonsubsampled contourlet transform
Al-Asad Medical ultrasound image: A pre-processing approach towards reconstruction
Birla et al. Performance analysis of image enhancement techniques for brain tumor images
Singh et al. Enhanced Steerable Pyramid Transformation for Medical Ultrasound Image Despeckling
Omar et al. Wavelet-based medical image fusion via a non-linear operator
CN118121233B (en) Ultrasonic microvascular imaging method and system based on singular value decomposition and Frangi filtering
Khan et al. Retinal image enhancement using Laplacian pyramidal multi-scaling

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant